Heuristics Work Until They Don’t

A short epistemics post against forming strong heuristics from a few salient cases. AI researchers learned 'progress is always slower than expected' from two AI winters, then got blindsided (a quick Bayesian calc puts ~33% on the next era surprising them — n=2 is weak). Romney 'unskewed' the polls and lost exactly as predicted, becoming a King-Canute parable about just believing the polls — until 2016, when the decreased-turnout story he'd invented actually came true (the n=1 parable was over-learned). And the Hillary-trusted-Big-Data-over-ground-game narrative teaches the exact opposite lesson from Moneyball. The point: 'stop treating life as a series of moral parables' — single cases are evidence, not lessons. Caps it with a self-referential joke: if he's convinced you not to over-learn from a few salient examples, 'then shame on you.'

Quality 67 · Strong. 67 — high-Strong. A tight, memorable epistemics post with a real insight (don't over-learn from n=1/n=2 salient cases; the Moneyball-vs-2016 inversion is genuinely illuminating) and a delightful self-undermining capstone. Slight in scope, which keeps it high-Strong.

Claude’s paradigm shift 48 · Moderate. 48 — Moderate. A fresh, memorable framing ('stop treating life as moral parables') of small-sample / base-rate reasoning.

Real-world impact 2 · Minor. 2 — a within-blog epistemics insight; no material reach.